Multi-Species Object Detection in Drone Imagery for Population Monitoring of Endangered Animals

Sowmya Sankaran

2024-06-28

Abstract:Animal populations worldwide are rapidly declining, and a technology that can accurately count endangered species could be vital for monitoring population changes over several years. This research focused on fine-tuning object detection models for drone images to create accurate counts of animal species. Hundreds of images taken using a drone and large, openly available drone-image datasets were used to fine-tune machine learning models with the baseline YOLOv8 architecture. We trained 30 different models, with the largest having 43.7 million parameters and 365 layers, and used hyperparameter tuning and data augmentation techniques to improve accuracy. While the state-of-the-art YOLOv8 baseline had only 0.7% accuracy on a dataset of safari animals, our models had 95% accuracy on the same dataset. Finally, we deployed the models on the Jetson Orin Nano for demonstration of low-power real-time species detection for easy inference on drones.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

This paper mainly discusses how to use unmanned aerial vehicle (UAV) imagery for multi-species animal detection to solve the problem of accurate counting of endangered species. Currently, the population of wild animals is declining rapidly, and accurate counting techniques are crucial for monitoring population changes. The study focuses on fine-tuning object detection models, particularly the YOLOv8 architecture for UAV images, to improve the accuracy of animal species recognition.

The authors trained 30 different models and applied hyperparameter optimization and data augmentation techniques to improve accuracy. Compared to the YOLOv8 baseline model with 0.7% accuracy on a wildlife dataset, their model achieves 95% accuracy. Additionally, they deployed the model on NVIDIA Jetson Orin Nano to achieve low-power real-time species detection, facilitating inference on UAVs.

The main contributions of the paper are as follows:
1. Fine-tuning the YOLOv8 architecture, resulting in a 135-fold improvement in recognition accuracy for multi-species UAV imagery.
2. Achieving up to 98.2% accuracy in recognizing UAV imagery containing ten different species through comprehensive hyperparameter optimization.
3. Deploying the model on edge GPU (Jetson Orin Nano) for low-power real-time object detection.

This research is of great significance for monitoring endangered species populations in wildlife conservation areas with limited resources, providing an accurate and cost-effective solution. In the future, the authors plan to expand the dataset to include more species and develop a fully automated detection system for UAV applications.